Slower downstream stage affects performance of upstream stag

Post questions here relative to DataStage Enterprise/PX Edition for such areas as Parallel job design, Parallel datasets, BuildOps, Wrappers, etc.

Moderators: chulett, rschirm, roy

Post Reply
chetan.c
Participant
Posts: 112
Joined: Tue Jan 17, 2012 2:09 am
Location: Bangalore

Slower downstream stage affects performance of upstream stag

Post by chetan.c »

Hi,

My job design looks like below.

Code: Select all

Ext_Source---transformer---Java Transaformer---Java Transaformer---Transformer--dataset.
When i run the Code only from

Code: Select all

External source --Transformer---Dataset
, the rows/second from the stages are around 1400rows/sec.

But if the Complete design is included the External source_stage and transformer output slows down to 250rows/sec!!.

Is my inference correct that a slower perforrming downstream stage affects faster performing upstream stages?


Thanks,
Chetan.C
chulett
Charter Member
Charter Member
Posts: 43085
Joined: Tue Nov 12, 2002 4:34 pm
Location: Denver, CO

Post by chulett »

Of course. Which is why they are known as bottlenecks.
-craig

"You can never have too many knives" -- Logan Nine Fingers
chetan.c
Participant
Posts: 112
Joined: Tue Jan 17, 2012 2:09 am
Location: Bangalore

Post by chetan.c »

Thanks Craig.
chulett
Charter Member
Charter Member
Posts: 43085
Joined: Tue Nov 12, 2002 4:34 pm
Location: Denver, CO

Post by chulett »

Identification of them is the first step, after that you can work on ameliorating them. What's going on in the Java transformers? Any way to replace them with native DataStage functionality?
-craig

"You can never have too many knives" -- Logan Nine Fingers
chetan.c
Participant
Posts: 112
Joined: Tue Jan 17, 2012 2:09 am
Location: Bangalore

Post by chetan.c »

chulett wrote: What's going on in the Java transformers? Any way to replace them with native DataStage functionality? ...
Hi Craig,

The java transformer has the code to generate XPATH,element name and value.We are using this as the xsd of the XML's that we get comes from over 1000 applications and all have their own structures.

The other java transformer does cleansing of XML.

So its not possible to parse even with XML assembly stage also.

That's why we had to go for the java code.

Thanks,
Chetan.C
chulett
Charter Member
Charter Member
Posts: 43085
Joined: Tue Nov 12, 2002 4:34 pm
Location: Denver, CO

Post by chulett »

Well... then it sounds like you're stuck with the neck of that particular bottle.
-craig

"You can never have too many knives" -- Logan Nine Fingers
jerome_rajan
Premium Member
Premium Member
Posts: 376
Joined: Sat Jan 07, 2012 12:25 pm
Location: Piscataway

Post by jerome_rajan »

The lag is being caused by the way buffers work in DataStage and is an essential.

You might want to have a look at the "score" to ensure that no unnecessary repartitioning is happening.

Also, if your Transformer is not implementing any business logic and is just performing minor conversions, you might want to have a look at the Modify stage.

If performance is a very critical facet and you have substantial memory at your disposal, try to increase the buffer size.

Not too sure if it would be a wise idea to increase the APT_BUFFER_FREE_RUN but it might help. Trial & error will give you a good idea
Jerome
Data Integration Consultant at AWS
Connect With Me On LinkedIn

Life is really simple, but we insist on making it complicated.
chetan.c
Participant
Posts: 112
Joined: Tue Jan 17, 2012 2:09 am
Location: Bangalore

Post by chetan.c »

chulett wrote:Well... then it sounds like you're stuck with the neck of that particular bottle. ...
:) yes right at the neck ..we are trying to improve performance of the Java code right now..

Hi Jerome,
Thanks . I did think about buffer memory but thought first let us tune the java code as much as possible and then look into that.

Thanks,
Chetan.C
Post Reply